Toward a Simonian Theory of Integration

Philippe Jehiel¤

January 2003

Abstract

From a positive viewpoint, business as usual decisions unlike other de-

cisions do not generally require further justi¯cations. And when a justi¯ca-

tion is required, decision makers are often judged by results: if the resulting

performance is above a pre-de¯ned threshold this is ¯ne; otherwise this is

considered a failure. This paper explores the cost and bene¯t of integration

under such a Simonian working of reviewing processes. We identify the fol-

lowing cost of non-integration : bargaining ine±ciencies arise because the

threshold performances that determine the criterion of success need not be

adapted to every single negotiation. And the main cost of integration is

its exposure to opportunism because the opportunism in one division has

spillover e®ects in the other divisions of integrated organizations.

¤CERAS, Paris and University College London. mailing address: C.E.R.A.S.-E.N.P.C.,

C.N.R.S. (URA 2036), 48 Bd Jourdan, 75014 Paris, France; e-mail: jehiel@enpc.fr.

1

1 Introduction

Many managerial decisions have e®ects far beyond the organization to which the

decision maker belongs. That is, there are externalities. In a zero-transaction cost

world the decision makers of the various concerned organizations would negotiate

to overcome these externalities.

But, managers generally have to justify at least some of their decisions to

the review boards in charge of assessing them. And review boards are generally

not aware of the details of the information available at the time of the decision

making. They need not even be aware that a negotiation took place. They are

thus forced to rely on more primitive instruments to assess their managers. In

particular managers are often "judged by results". That is, if the decision results

in a satisfactory performance the decision is considered successful. Otherwise it

is a failure.

We will illustrate how bargaining ine±ciencies may arise in such contexts. Of

course, ine±ciencies will not always arise, and sometimes e±cient negotiations

will e®ectively take place.1 More precisely, bargaining ine±ciencies will arise only

when the gains induced by the negotiation are not high enough relative to the

benchmark performances that de¯ne the managerial criterion of success. In such

scenarios, no negotiation takes place and suboptimal decisions are being made.

The fundamental reason for bargaining ine±ciencies here is that the bench-

mark performances against which managerial decisions are assessed are not (solely)

determined by the performances that would arise in the speci¯c decision process

under review if no negotiation were to take place.2 To put it di®erently, bar-1This should be contrasted with frameworks in which no negotiation whatesoever can take

place (see Hart and Holmstrom 2002 for a recent theory of the ¯rm in a such a setup). Then

ine±ciencies would arise as soon as the private interest of the decision maker would diverge

from the collective interest. That is, as soon as a negotiation would be needed.2In case a negotiation would take place, the review boards would not have access to the

performances that would have arisen in the absence of negotiation.

2

gaining ine±ciencies arise because the benchmark performance used to assess the

management is the same for a range of decisions, and thus it is not always ¯tted

to the speci¯c decision process under review.

This paper is an attempt to bridge together two in°uential lines of thoughts in

the theory of organization that appear to have had little in°uence on each other.

On the one hand, organization theorists in the tradition of Coase and Williamson

(1975) have been developing theories that aim to explain how large ¯rms should

be (see Grossman and Hart (1986) for a prominent theory along these lines). One

of the main themes in this tradition is about the costs and bene¯ts of integration.

On the other hand, organization theorists in the tradition of March and Simon

(1958) have been stressing that decision making in organizations follow patterns

that radically di®er (by their simplicity) from how rational behavior is usually

modeled in economics.

The above description of the functioning of review boards follows the tradi-

tion of March and Simon in that the benchmark performances which de¯ne the

criterion of success play much the same role as the aspiration levels (and associ-

ated satis¯cing behaviors) in the theory of Simon (1955). And, in the tradition of

Coase and Williamson our main interest lies in shedding light on the cost and ben-

e¯t of integration as generated by such a functioning of review boards, in a world

where managers can either be dedicated to their organization or opportunistic.

As already stressed, a key ingredient of the model we propose is the working of

the review when it takes place. Another key ingredient is about the speci¯cation

of when reviews take place. Obviously, in the real world not every single decision

triggers a review. (This would be too costly and/or time consuming for the review

board.) Some decisions are reviewed whereas others are considered as acceptable

ones without further justi¯cation.3

3We implicitly assume in the analysis to be presented below that the decision to make a

review can only be based on the decisions made by the manager, not on the performance

resulting from these decisions. The rationale is that unless a review takes place it is hard to

3

In our analysis, benchmark performances and acceptable decisions are related

as follows. In any organization, the benchmark performance that de¯nes the

managerial criterion of success is assumed to coincide with the average perfor-

mance obtained over all scenarios in which acceptable decisions are being made.

Thus, in any given organization either an acceptable decision is being made - in

which case on average the benchmark performance is being obtained - or another

decision (possibly involving side-payments and negotiations) is being made - in

which case the working of the review forces the organizational performance to

be no smaller than the benchmark performance (whether or not the manager is

opportunistic). Overall, given the working of the organization the performance is

bound to be no smaller than the threshold performance. Thus, in Simonian terms

the system is likely to be stable to the extent that the threshold performance is

satisfactory to the board.4

In the rest of the paper we develop a theory of integration based on the above

ingredients. We have mentioned above the bargaining ine±ciencies arising in

non-integrated structures. By contrast, integrated organizations do not give rise

to bargaining ine±ciencies. However, in integrated structures, when the manager

of one division is opportunistic it may divert the surplus generated in another

division regardless of whether the manager of the latter division is dedicated or

opportunistic. Thus, the opportunism of managers within an integrated organi-

disentangle the e®ects of the various decisions contributing to the overall performance.4In our analysis the sets of acceptable decisions - for which there is no review - are set

exogenously. One possible interpretation is that acceptable decisions are "business as usual

decisions", and the review boards would ¯nd it too costly to review systematically every such

decision. Another (more ambitious) interpretation is that the set of acceptable decisions is the

outcome of a search process on the part of the review board. (That is, in each organization the

review board will adjust its set of acceptable decisions until a decent performance is obtained.)

We do not model the search process, and we assume that every organization has reached a

point that is satisfactory to the board members. (This echoes considerations appearing in

Simon (1955).)

4

zation is a public bad among the divisions of the organization.

To summarize, the main cost of non-integration lies in the bargaining ine±-

ciencies arising when the surplus generated by the negotiation is not high enough

relative to the benchmark performances that de¯ne the criterion of success in the

various organizations. And the main cost of integration lies in the public bad

aspect of opportunism within organizations.

Thus, in a world where either there are no externalities or very severe ex-

ternalities (but no intermediate externalities) between the activities of various

divisions integrating the various divisions is not desirable. And in a world with

very little opportunism integration dominates non-integration.

The theory is admittedly highly stylized, but we believe it captures impor-

tant elements of the cost/bene¯t analysis of integration/decentralization. For

example, France has had a long tradition of highly centralized administration.5

Beyond the political pressure from the regions that led to criticize centralization,

it is generally agreed that the French Administration was relatively e±cient, as

long as civil servants had a su±ciently strong feeling for the so-called "Service

Public", i.e. as long as civil servants were su±ciently dedicated to the com-

mon interest in our terminology.6 Our theory agrees with this, as it predicts

that whenever decision makers are dedicated, centralized/integrated organiza-

tions dominate decentralized/non-integrated ones. It may also explain why in

countries like the US in which there is no strong tradition for the "Service Pub-

lic" a decentralized administration may be preferable.

In a di®erent vein, many economists (in the tradition of Pigou) would see

the presence of externalities as an argument in favor of integrating/centralizing

the decision making. But, there are many examples in which obvious (and well5It has only recently since Deferre's 1982 laws evolve toward a more decentralized

organization.6It is also generally considered that the feeling for "Service Public" in the French adminis-

tration has signi¯cantly deteriorated over the past two decades.

5

identi¯ed) divergences of interest have led the concerned parties to negotiate so

that no signi¯cant ine±ciencies occurred despite the decentralized aspect of the

decision making. To take a recent example, the refugee camp of Sangatte located

close to the Chunnel tunnel was clearly a source of divergence of interest between

the French and the British. Despite the strong divergence and the decentralized

decision making (the EU had no say in this issue), an agreement about how to

distribute the refugees between France and the UK was easily found. Hence,

despite the divergence of interest, a decentralized decision making seems to have

worked e±ciently in this case. Our theory is also consistent with this, since in

our approach decentralized decision making is mostly problematic whenever the

externalities are not too severe, i.e. whenever the divergence of interest is not

too strong (in the Sangatte case, the divergence of interest was very clear and

strong).

Section 2 develops the model and describes the decentralized and the inte-

grated structures. Section 3 discusses the issue of bargaining ine±ciencies. Sec-

tion 4 discusses the cost of integration. Further insights and comparative statics

with respect to the set of acceptable decisions and threshold performances appear

in Section 5. The relationship with the literature is discussed in Section 5.

2 The Model

Consider the manager M of an organization O. Decisions must be made, which

include physical decisions (like investment decisions), and also possibly side-

payments or monetary transfer decisions to or from other organizations (or in-

terests).

An important aspect of this paper is that decisions are categorized into two

classes: "business as usual" decisions, and other decisions. A key assumption is

that "business as usual" decisions do not require further justi¯cations whereas

6

any other decision need to be justi¯ed.7 When a justi¯cation is required, there

is a review: the performance ¼O of the decision under review is then made clear,

and the manager is assumed to be "judged by results". That is, if ¼O turns out to

be above some pre-speci¯ed threshold performance ¼TO, the decision is justi¯ed,

but not otherwise.8

We will assume that the threshold performance ¼TO coincides with the average

performance obtained when "business as usual" decisions are being made. Thus,

whatever the environment (that the review board of organization O need not

even be aware of), the procedure is bound to produce an expected performance

no smaller than ¼TO.9010

We will further assume that all managers care (a lot) about not making de-

cisions that would trigger a review and that could not be justi¯ed. The idea

is that such decisions would hurt managers' career, and thus all managers will

try their best to avoid making decisions that cannot be justi¯ed ex post. Apart7Business as usual decisions are assumed to be known to deliver an acceptable performance

on average- this is supposed to have been learnt from past experiences. This in turn explains why

such decisions are considered as acceptable ones without further justi¯cation. The performance

of any other decision however is not accessible (not even probabilistically, as would argue non-

Bayesian organization theorists such as March and Simon 1955) to those in charge of assessing

the manager.8The idea that decisions are solely judged by results is clearly extreme. It ¯ts with situations

in which other aspects of the decisions (like the deliberations leading to the decision making)

are not accessible to the review board (possibly because the review board has no time to study

whether arguments are well funded or not).9To the extent that ¼T

O is an acceptable performance, the above procedure meets the sat-

is¯cing requirement of Simon (1955) regardless of the environment. We do not model the

search process that has led to the choice of acceptable decisions and its derived threshold per-

formance. This should be the subject of future research. Implicitly (in accordance with our

chosen wording), we assume here that acceptable decisions are the more familiar ones.10This robustness with respect to the environment may explain why such procedures are so

commonly used in practice. They need not be optimal however in speci¯c environments (for a

discussion about good versus optimal procedures, see Simon 1955).

7

from their common desire not to make decisions that cannot be justi¯ed ex post,

managers can be of two possible types: either they are dedicated to their organi-

zation or they are opportunistic. When dedicated to his organization, a manager

cares about the performance of his organization. When opportunistic, a manager

cares about his own well-being. Manager M is dedicated with probability ¸M

and opportunistic with probability 1 ¡ ¸M .

The goals of dedicated versus opportunistic managers are highly stylized (and

thus they are not meant to be realistic). This modeling should be thought of as

providing a stylized representation of the potential non-congruence between the

managerial and the organizational interest.11

The main objective of this paper is to analyze the cost and bene¯t of inte-

gration in an organizational setup such as the one described above. We will in

turn analyze a decentralized structure and an integrated structure, and we will

make comparative statics with respect to the economic performance of the two

structures.

In the decentralized structure, two managers i = 1; 2 run separate organiza-

tions i = 1; 2. They each make decisions that can induce externalities on the other

organization. But, they are allowed to bargain to overcome these externalities.

In the integrated structure, the (same) two managers i = 1; 2 belong to the

same organization 0 and the same decisions as in the decentralized case have to be

made by each manager. Of course, when dedicated, manager i's objective is now

the performance of the integrated organization 0 because this is the organization

to which he belongs. To perform the comparison between the decentralized and

integrated structures, we will make the (natural) assumption that the business11Issues about how to align the interest of the manager with that of the organization through

an appropriate choice of wage contract is not the subject of this paper. Implicitly, we adopt an

adverse selection perspective in which it is assumed that within the range of admissible wage

contracts, the type of the manager is una®ected (or in a moral hazard formulation, he has no

su±cient incentives to change his goal).

8

as usual decisions in organization 0 correspond to the pro¯le of business as usual

decisions in organizations 1 and 2 of the decentralized structure. In the integrated

organization as in any organization, any decision that is not "business as usual"

needs justi¯cation, and the management is then judged by results, as explained

above. When a decision fails to be justi¯ed, we assume that there is a prejudice to

the entire management, that is, to both managers i = 1; 2. Thus, both managers

i = 1; 2 will try their best to avoid making decisions that cannot be justi¯ed ex

post from the viewpoint of the integrated organization 0.

In the next two subsections, we describe in more detail the working of the

decision making in the two modes of organization.

2.1 Decentralized structure

The state and decision spaces:

We let µ 2 £ denote the state of the world. The physical decision in organiza-

tion i is denoted di, the transfer from inside to outside organization i is denoted ti

where ti = t¡ii +tmi , t¡ii stands for the transfer from organization i to organization

¡i and tmi (¸ 0) stands for the transfer from organization i to manager i.12 Thus,

a decision in organization i can be seen as a triple ai = (di; tmi ; t¡ii ).

The physical decision di belongs to Ai [ Ji where Ai is the set of business as

usual decisions and Ji is the set of other decisions.

Organizational payo®:

We let

¼i ´ ui(di; d¡i; µ) ¡ ti

denote the performance of organization i in state µ when the physical decisions

are given by (di; d¡i) and the transfer from organization i (to other interests) is

given by ti.

12tmi > 0 will possibly arise due to the opportunism of manager i.

9

Monitoring technology:

The review board in i freely observes di, ti, and it observes the realized per-

formance ¼i in case of review; nothing else.13 If di 2 Ai and ti = 0, (di; tmi ; t¡ii )

is an acceptable decision for organization i, and there is no need for further jus-

ti¯cation. Otherwise, i.e. if di =2 Ai or if ti 6= 0, the decision ai = (di; tmi ; t¡ii )

must be justi¯ed. As explained above, ai is justi¯ed whenever ¼i ¸ ¼Ti . It is notjusti¯ed otherwise.

Managerial information and payo®:

At the time the decisions are to be made, managers are assumed to know

the state of the world µ, the actions spaces in both organizations Ai and Ji,

i = 1; 2, and the functional form of the performance ui in both organizations.

When manager i is dedicated his payo® is given by:

vDedi (ai; a¡i; µ) ´

8><>:¼i if ¼i ¸ ¼Ti or (di 2 Ai and ti = 0)

¡1 otherwise

When manager i is opportunistic his payo® is given by:

vOppi (ai; a¡i; µ) ´

8><>:tmi if ¼i ¸ ¼Ti or (di 2 Ai and ti = 0)

¡1 otherwise

We allow for negotiations about (ai; a¡i) between managers i and ¡i. The out-

come of the negotiation is assumed to be Pareto-e±cient given the managerial

payo® functions. It should also Pareto-dominate the Non-cooperative outcome

(or one of them if there are several of them). It will be described in more details

through examples later on. For the time being, note that the transfers tmi ; t¡ii ,

13In particular, it is incapable of observing t¡ii and tmi separately (nor does it know a priori

the other organizations with which i may interact). It does not even need to know the set Ji.

10

i = 1; 2 must satisfy two feasibility constraints:14

t¡ii + ti¡i = 0

tmi + tm¡i + ¼i + ¼¡i = ui(di; d¡i; µ) + u¡i(d¡i; di; µ)

Acceptable decisions and threshold criterion:

The value of ¼Ti is determined by:

¼Ti =Eµ[ui(di(µ); d¡i(µ); µ) j di(µ) 2 Ai and ti(µ) = 0]

where dj(µ) (resp. ti(µ)) denotes the equilibrium physical decision in organization

j (resp. transfer from organization i) in state µ. That is, the threshold criterion

coincides with the expected equilibrium value of organization i 's performance

conditional on an acceptable decision (di; tmi ; t¡ii ), di 2 Ai, ti = 0 being made in

organization i.

2.2 Integrated structure

The state and decision spaces:

The state and decision spaces are the same as in the decentralized structure.

That is, manager i's action is ai = (di; tmi) where di is the physical decision in

division i and tmi is the transfer from the integrated organization 0 to manager

i.15

14The ¯rst condition expresses an accounting constraint (what goes from i to ¡i should be the

opposite of what goes from ¡i to i). The second condition expresses that the money distributed

among the organizations and the managers has to come from payo®s produced by the physical

decisions.15There is no transfer to organizations outside 0 because we assume there are no externalities

with these. (However, it is important to keep in mind that the review board need not be aware

of that, so that it cannot regard the fact that there is a transfer from organization 0 as a sign

that his management is corrupt.)

11

Organizational payo®:

It is given by:

¼0 ´ ui(di; d¡i; µ) + u¡i(d¡i; di; µ) ¡ tmi ¡ tm¡i

Monitoring technology:

The review board in 0 observes d1, d2, t = tmi + tm¡i , and it observes the

realized performance ¼0 in case of review. If di 2 Ai i = 1; 2 and t = 0,

(d1; d2; tm1 ; tm2) is an acceptable decision for organization 0 and there is no need

for further justi¯cation. Otherwise, i.e. if di 2 Ji for i = 1 or 2, or if t 6= 0,

the managerial joint decision a = (d1; d2; tm1; tm2) must be justi¯ed; a is justi¯ed

whenever ¼0 ¸ ¼T0 . It is not justi¯ed otherwise.

Managerial information and payo®:

When manager i is dedicated his payo® is given by:

vDedi (ai; a¡i; µ) ´

8><>:¼0 if ¼0 ¸ ¼T0 or [(d1; d2) 2 A1 £A2 and t = 0]

¡1 otherwise

When manager i is opportunistic his payo® is given by:

vOppi (ai; a¡i; µ) ´

8><>:tmi if ¼0 ¸ ¼T0 or [(d1; d2) 2 A1 £ A2 and t = 0]

¡1 otherwise

Our speci¯cation assumes that both managers i = 1; 2 get ¡1 whenever a

decision other than business as usual is being made and the resulting performance

falls short of the threshold performance ¼T0 . This is consistent with the view that

based on what the review board observes it is unable to distinguish who in the

organization might be responsible for the bad performance.

Acceptable decisions and threshold criterion:

The value of ¼T0 is determined by:

¼T0 =Eµ[ui(di(µ); d¡i(µ); µ)+u¡i(d¡i(µ); di(µ); µ) j dj(µ) 2 Aj, j = 1; 2 and t(µ) = 0]

where dj(µ) (resp. t(µ)) denotes the equilibrium physical decision of division j

(resp. transfer from organization 0) in state µ.

12

3 On Bargaining Ine±ciencies

An interesting implication of the setup is that - in the decentralized setting -

the decisions may be ine±cient even though we allow for bargaining between the

managers. As we will show, managers bargain less often than what would be

optimal (from the overall organizational viewpoint) because they are not willing

to make decisions that cannot be justi¯ed ex post. Our paper thus points out

to a new form of transaction costs (Coase 1960) that we believe is of practical

major importance.

It is worth mentioning that the bargaining ine±ciency does not arise because

of the possibility that managers are opportunistic. As it turns out the same

physical decisions are made irrespective of whether managers are dedicated or

opportunistic.16 Bargaining ine±ciencies arise because managers have to justify

their "non-business-as-usual" decisions, and the justi¯cation technology is based

on the "judged by results" principle.

To illustrate the claim, consider the following simpli¯cation:

Manager 1 must make a physical decision d1 in fdA; dJg, where dA denotes the

business as usual decision and dJ requires further justi¯cation, i.e., A1 = fdAgand J1 = fdJg. Manager 2 has no physical decision to make (i.e., A2 = J2 = ;).Decision dA induces a nul payo® both in organizations 1 and 2, irrespective of

the state of the world µ. That is,17 for all states µ,

ui(dA; µ) = 0:

An interpretation is that dA corresponds to the status quo decision, and the16The type of the manager is however a key determinant with respect to the distribution of

the rent created by the decisions. A dedicated manager leaves the entire rent to his organization

whereas an opportunistic manager takes the maximal rent for himself leaving only the minimum

threshold performance to his organization.17Since there is no physical decision in organization 2, payo®s depend solely on the decision

d1 and the state of the world µ. Notations are simpli¯ed accordingly.

13

status quo delivers a constant payo® (normalized here to 0) to all concerned

organizations whatever the state of the world.

For the following discussion, it is convenient to distinguish states according

to whether decision dJ is preferable to dA from the viewpoint of organization 1

and from the aggregate organizational viewpoint. This leads us to divide states

of the world into four categories. Denoting by u(d1; µ) = u1(d1; µ) + u2(d1; µ) the

aggregate organizational payo®, we let:

£CongJ = fµ j u1(dA; µ) < u1(dJ ; µ) and u(dA; µ) < u(dJ ; µ)g

£CongA = fµ j u1(dA; µ) ¸ u1(dJ ; µ) and u(dA; µ) ¸ u(dJ ; µ)g

£DivJ = fµ j u1(dA; µ) ¸ u1(dJ ; µ) and u(dA; µ) < u(dJ ; µ)g

£DivA = fµ j u1(dA; µ) < u1(dJ ; µ) and u(dA; µ) ¸ u(dJ ; µ)g

£Cong ´ £CongA [ £CongJ refers to those states in which the private interest

of organization 1 coincides with the aggregate organizational interest. £Div ´£DivA [ £DivJ refers to those states in which the private interest of organization 1

diverges from the aggregate organizational interest. £A ´ £CongA [ £DivA denotes

those states for which dA is best from the aggregate organizational viewpoint.

£J ´ £CongJ [ £DivJ denotes those states for which dJ is best from the aggregate

organizational viewpoint.

Preliminary observations:

1) The threshold performance in organization 1 is ¼T1 = 0. This because

u1(dA; µ) = 0 for all µ, and the relation of the threshold performance to the

organizational payo®.

2) Whenever there is congruence between the private interest of organization

1 and the aggregate organizational interest (i.e. µ 2 £Cong) the e±cient decision

is made, and there is no negotiation. That is, d1(µ) = dA if µ 2 £CongA and

d1(µ) = dJ if µ 2 £CongJ .18

18This is clearly true whenever µ 2 £CongA . Decision dJ is chosen whenever µ 2 £Cong

J because

14

Observe also that when µ 2 £Cong and u2(d1; µ) > ¼T2 , manager 2 when

opportunistic diverts tm2 = u2(d1; µ) ¡ ¼T2 from organization 2.

3) For µ 2 £DivJ , the e±cient decision d1(µ) = dJ arises (through bargaining)

whenever19

u(dJ ; µ) ¸ ¼T2

Otherwise, there is no bargaining and the ine±cient decision d1(µ) = dA is chosen.

Observe that if ¼T2 · 0, there is always an e±cient decision because u(dJ ; µ) >

u(dA; µ) = 0 for µ 2 £DivJ .

4) For µ 2 £DivA , the e±cient decision d1(µ) = dA arises (through bargaining)

whenever

0 ¸ u1(dJ ; µ) + ¼T2

Otherwise, there is no bargaining and the ine±cient decision d1(µ) = dJ is chosen.

The bargaining ine±ciencies arising for µ 2 £Div can be described as follows.

For the sake of illustration, consider µ 2 £DivA and suppose that u1(dJ ; µ)+¼T2 > 0.

Here, no bargaining takes place in equilibrium and the resulting physical decision

is dJ . This is ine±cient from the aggregate organizational viewpoint because by

de¯nition we have that u(dJ ; µ) < u(dA; µ) for µ 2 £DivA . In a zero-transaction

cost world, such an ine±cient decision could not be the ¯nal outcome. Indeed,

manager 1 might instead propose the following bargain to manager 2: "I will make

decision dA (rather than decision dJ) if you, manager 2, transfer an amount of

money of t12 = u1(dJ ; µ)+u(dA;µ)¡u(dJ ;µ)

2 to organization 1." If such a deal were to be

accepted by manager 2, manager 1 would be happy with it in our setup. However,

manager 2 would not accept such a deal. The problem is that by so doing the

net payo® of organization 2 would only be u2(dA; µ) ¡ t12 = ¡u1(dJ ; µ) + u(dJ ;µ)2 ,

and this payo® is strictly less than ¼T2 (remember that u(dJ ; µ) < u(dA; µ) = 0).

Hence, manager 2 (whether dedicated or not) would end up making a decision

u1(dJ ; µ) > u1(dA; µ) = 0 and ¼T1 = 0, see above.

19This is because the threshold performance in organization 1 is 0, see above.

15

that is not business as usual (since t12 > 0) and that could not be justi¯ed ex post

(since it would result in a payo® that is inferior to the threshold payo® ¼T2 ). Thus,

manager 2 would refuse such a deal, and more generally no bargaining (however

the surplus is shared between the parties) can take place in such a case.

It is worth stressing that the source of the problem is that manager 2 is not

evaluated relative to the situation that would arise if20 no bargaining were to take

place;21 he is evaluated relative to the benchmark performance ¼T2 and ¼T2 is not

solely determined by this situation.

5) Finally, the threshold payo® in organization 2 satis¯es the following ¯xed

point condition:

¼T2 = ¸2Ehu2(d1(µ); µ) j µ 2 NBDed(¼T2 )

i+(1¡¸2)E

hu2(d1(µ); µ) j µ 2 NBOpp(¼T2 )

i

(1)

where

NBDed(¼T2 ) =nµ 2 £Cong or (µ 2 £DivJ and u(dJ ; µ) < ¼T2 ) or (µ 2 £DivA and u1(dJ ; µ) + ¼T2 > 0)

o

denotes the set of states in which there is no bargaining between the two organizations,22

and

NBOpp(¼T2 ) =

8><>:

(µ 2 £Cong and u2(d1(µ); µ) · ¼T2 ) or(µ 2 £DivJ and u(dJ ; µ) < ¼T2 ) or (µ 2 £DivA and u1(dJ ; µ) + ¼T2 > 0)

9>=>;

denotes the set of states in which there is no transfer from organization 2 (either

to organization 1 or to manager 2) whenever manager 2 is opportunistic.

Thus, ¼T2 is the expected value of u2(d1(µ); µ) conditional on no transfer being

made from organization 2 (either to organization 1 or to manager 2 and irre-

20The review board does not have access to such counterfactual data.21Here the no-negotiation payo® of organization 2 is u2(dJ ; µ), and such a benchmark payo®

would allow the managers to bargain e±ciently.22This is also the set of states in which there is no transfer from organization 2 whenever

manager 2 is dedicated.

16

spective of whether or not such transfers are the result of bargaining between

organizations 1 and 2).23

It is worth noting that ¼T2 · Ehu2(d1(µ); µ) j µ 2 NBded(¼T2 )

ifor all ¸2.

Bargaining ine±ciencies and divergence of interest:

We start by assuming that u2(dJ ; µ) = 0 for all µ 2 £CongJ . That is, when the

interest of organization 1 is congruent with the aggregate organizational interest,

the payo® to organization 2 is 0. In this case, we have:

Proposition 1 Suppose that u2(dJ ; µ) = 0 for all µ 2 £CongJ . Then (1) An

ine±cient decision must occur for some µ 2 £DivA ; (2) The e±cient decision

always arises whenever µ 2 £DivJ . And this holds true whatever the speci¯cations

of ¸1, ¸2, the distributions of payo®s and of the states of world.

Proof. Observe ¯rst that ¼T2 · 0. (This is because ¼T2 is a convex combination

of u2(dA; µ) = 0 and of u2(dJ ; µ) for µ 2 £DivA , u2(dJ ; µ) = 0 for µ 2 £CongJ and

u2(dJ ; µ) < 0 whenever µ 2 £DivA .) This implies (2) that bargaining is always

e±cient whenever µ 2 £DivJ (see the third preliminary observation). Suppose

now (by contradiction to 1) that the decision is always e±cient whenever µ 2£DivA . We should then have that ¼T2 = 0 (because NBDed(¼T2 ) would not contain

any µ 2 £DivA ). But, for all µ 2 £DivA , we have that u1(dJ ; µ) > 0. Hence,

u1(dJ ; µ) + ¼T2 > 0 and there is no room for an e±cient bargaining whenever

µ 2 £DivA , thus yielding a contradiction.

Beyond showing the presence of bargaining ine±ciencies (for some µ 2 £DivA ),

the above Proposition also shows that it is not the mere divergence of interest

between the two organizations that causes the ine±ciency. Indeed, ine±ciencies

must occur for some µ 2 £DivA , but never for µ 2 £DivJ . So the mere divergence

of interest is not an indication as to whether or not an ine±ciency must occur.23Observe that ¼T

2 does not depend on the probability ¸1 that manager 1 is dedicated. This

is because the type of manager 1 has no e®ect on whether or not there are transfers from

organization 2 in equilibrium.

17

To generalize Proposition 1, we allow for more general speci¯cations of u2(dJ ; µ),

µ 2 £CongJ . As the following Proposition shows, e±cient bargaining takes place

for all µ 2 £DivJ in a number of situations:

Proposition 2 There always exists ¸2 such that for all ¸2 · ¸2, there is an

e±cient bargaining for all µ 2 £DivJ . Besides, if Ehu2(dJ ; µ) j µ 2 £CongJ

i· 0,

the decision is always e±cient for all µ 2 £DivJ and whatever ¸2.

Proof. As for the ¯rst part, observe that ¼T2 must be non-positive as ¸2 ap-

proaches 0. (This is because as manager 2 gets opportunistic with a probability

close to 1, he gets to divert some positive amount from his organization whenever

the realization of u2(dJ ; µ), µ 2 £CongJ is above ¼T2 , which can be made consistent

only if ¼T2 · 0, see expression (1).) Part 1 follows, since ¼T2 · 0 implies that there

is e±cient bargaining for µ 2 £DivJ whenever ¼T2 · 0. As for the second part, ob-

serve that for all ¸2, ¼T2 is no greater than a convex combination of u2(dA; µ) = 0

and Ehu2(dJ ; µ) j µ 2 £CongJ

i. (For µ 2 £DivA the absence of bargaining implies

that u2(d1; µ) < 0.) Hence, ¼T2 · 0 if Ehu2(dJ ; µ) j µ 2 £CongJ

i· 0, and we may

conclude.

The following Proposition identi¯es conditions under which bargaining ine±-

ciencies must occur for some µ 2 £DivA :

Proposition 3 Let w2 = minµ2£CongJ

u2(dJ ; µ). Suppose that

maxµ2£DivA

u1(dJ ; µ) + min(w2; 0) > 0.

Then whatever ¸2 bargaining ine±ciencies must arise for some µ 2 £DivA .

Proof. Observe that if bargaining is e±cient for µ 2 £DivA we must have ¼T2 ¸min(w2; 0). But, max

µ2£DivAu1(dJ ; µ) + ¼T2 ¸ max

µ2£DivAu1(dJ ; µ) + min(w2; 0) > 0 implies

that bargaining cannot be e±cient for all µ 2 £DivA .

Corollary 1 Suppose that w2 = minµ2£CongJ

u2(dJ ; µ) ¸ 0. Ine±cient bargaining must

occur for some µ 2 £DivA .

18

Proof. min(w2; 0) = 0 and for all µ 2 £DivA , u1(dJ ; µ) > 0. Hence, maxµ2£DivA

u1(dJ ; µ) + min(w2; 0) > 0.

In contrast with Proposition 1, we now show conditions under which bargain-

ing ine±ciencies do arise for some µ 2 £DivJ :

Proposition 4 Suppose that £DivA = ;. Assume that Ehu2(d1(µ); µ) j µ 2 £Cong

i>

0. If

minµ2£DivJ

u(dJ ; µ) < Ehu2(d1(µ); µ) j µ 2 £Cong

i,

then an ine±ciency must arise for some µ 2 £DivJ whenever ¸2 is su±ciently

large.

Proof. Let ¸2 = 1. If bargaining is always e±cient whenever µ 2 £DivJ one must

have ¼T2 = Ehu2(d1(µ); µ) j µ 2 £Cong

i. But, min

µ2£DivJu(dJ ; µ) < E

hu2(d1(µ); µ) j µ 2 £Cong

i

implies that ine±ciencies must occur for some µ 2 £DivJ . More generally, ine±-

ciencies must arise for some µ 2 £DivJ whenever ¸2 is su±ciently large.

It should be noted that if minµ2£DivJ

u(dJ ; µ) > z2 (and £DivA = ;), then there

is no bargaining ine±ciency for µ 2 £DivJ . Since a larger u(dJ ; µ), µ 2 £DivJindicates a higher divergence of interest between organization 1 and the aggregate

organizational viewpoint, it follows that more divergence of interest may help the

functioning of the non-integrated structure in our setup.

4 The cost of integration

The main cost of integration in our setup is that one opportunistic manager is

enough to have a very negative e®ect on the functioning of the overall integrated

organization. More precisely, as soon as one manager is opportunistic, and even

if the other manager is dedicated, the surplus wherever it is created (in the

integrated organization) will be diverted by the opportunistic manager(s).

To illustrate the claim, assume ¯rst in the model of Section 2 that ui(di; d¡i; µ)

is independent of d¡i for all di and µ. (With some abuse of notation, we will write

19

ui(di; µ):) Manager i must make a physical decision in fdiA; diJg where diA denotes

the business as usual decision in division i. Assume further that ui(dAi ; µ) = 0 for

all µ. We refer to such a speci¯cation as the no interaction scenario.

Proposition 5 In the no interaction scenario, the non-integrated structure dom-

inates the integrated structure. More precisely, the net expected gain of non-

integration over integration is

¸1(1 ¡ ¸2)E[max(u1(d1J ; µ); 0)] + ¸2(1 ¡ ¸1)E[max(u2(d2J ; µ); 0)]:

Proof. All threshold payo®s must be 0 under the assumed speci¯cation. It

follows that the net expected payo® of organization i in the non-integrated struc-

ture is ¸iE[max(ui(diJ ; µ); 0)]. The net expected payo® of organization 0 in the

integrated structure is

¸1¸2E[max(u1(d1J ; µ); 0)] + ¸1¸2E[max(u2(d2J ; µ); 0)]:

The result follows.

To get further insights as to the comparison between the integrated and non-

integrated structures, consider the setup of Section 3 and assume that £DivA = ;,but £DivJ 6= ; (so unlike in the above Proposition the interest of organization 1 is

not always aligned with the aggregate organizational interest). Assume further

that both the interests of organization 1 and of organization 2 are aligned with

the aggregate organizational interest whenever µ 2 £Cong. That is, u2(dJ ; µ) ¸ 0

for all µ 2 £CongJ .

When manager 1 is opportunistic for sure, we have:

Proposition 6 Suppose that £DivA = ; and u2(dJ ; µ) ¸ 0 for all µ 2 £CongJ .

Non-integration dominates integration whenever manager 1 is opportunistic with

probability 1, i.e. ¸1 = 0. The expected gain of non-integration over integration

is no smaller than

¸2E[u2(dJ ; µ) j µ 2 £CongJ ] Pr(µ 2 £CongJ ):

20

Proof. The expected organizational payo® in the integrated structure is 0 (since

¼T0 = 0). In the non-integrated structure, we have that ¼T2 ¸ 0 (because £DivA = ;and u2(dJ ; µ) ¸ 0 for all µ 2 £CongJ ). For µ 2 £DivJ either no negotiation is

possible (because u(dJ ; µ) < ¼T2 ) - and then the organizational payo® is 0 - or

negotiation can take place and then there is at least a non-negative gain of ¼T2of the nonintegrated structure over the integrated one. Thus, when µ 2 £DivJ ,

the overall organizational payo® can be larger and is never smaller in the non-

integrated structure than in the integrated one. For µ 2 £CongJ , there is a zero

payo® in organization 1 as manager 1 is assumed to be opportunistic. When

manager 2 is opportunistic, the payo® in organization 2 is min(u2(dJ ; µ); ¼T2 ) ¸ 0.

When manager 2 is dedicated, the payo® in organization 2 is u2(dJ ; µ). Hence,

the expected gain of the non-integrated structure over the integrated one is no

smaller than ¸2E[u2(dJ ; µ) j µ 2 £CongJ ] Pr(µ 2 £CongJ ).

Under the same conditions, when manager 2 is opportunistic for sure, we

have:

Proposition 7 Suppose that £DivA = ; and u2(dJ ; µ) ¸ 0 for all µ 2 £CongJ .

Non-integration dominates integration whenever manager 2 is opportunistic with

probability 1, i.e. ¸2 = 0. The expected gain of non-integration over integration

is no smaller than

¸1E[u1(dJ ; µ) j µ 2 £CongJ ] Pr(µ 2 £CongJ )

and it is no greater than

¸1E[u1(dJ ; µ) j µ 2 £CongJ ] Pr(µ 2 £CongJ ) + ¸1E[u(dJ ; µ) j µ 2 £DivJ ] Pr(µ 2 £DivJ ):

Proof. The expected organizational payo® in the integrated structure is 0 (since

¼T0 = 0). In the non-integrated structure, we have that ¼T2 = 0 (because £DivA =

; and u2(dJ ; µ) ¸ 0 for all µ 2 £CongJ ). The expected overall organizational

surplus in the non-integrated case generated by organization 1 when µ 2 £CongJ

21

is ¸1E[u1(dJ ; µ) j µ 2 £CongJ ] Pr(µ 2 £CongJ ) (it corresponds to manager 1 being

dedicated). When manager 1 is dedicated and has the entire bargaining power,

the surplus generated for µ 2 £DivJ is ¸1E[u(dJ ; µ) j µ 2 £DivJ ] Pr(µ 2 £DivJ ).

Hence, the result.

Remark: Whenever manager 1 has some bargaining power (in the negotia-

tions arising for µ 2 £DivJ ), the gain of non-integration over integration positively

depends on E[u(dJ ; µ) j µ 2 £DivJ ]. Hence, "more" divergence of interest (or

an increase of £DivJ ) between organization 1 and the aggregate interest tends to

increase the relative advantage of non-integration over integration. This arises

because unlike in setups in which negotiations are not allowed, here negotiation

does take place whenever µ 2 £DivJ . And some of the surplus generated by such

negotiations bene¯ts organization 1 (in the integrated structure the surplus is

diverted by manager 2 who is assumed to be opportunistic with probability 1).

As the above results show, non-integration may dominate integration even

if the interest of organization 1 is not congruent (always) with the aggregate

organizational interest. When both managers are known to be dedicated for sure

however, integration dominates non-integration:

Proposition 8 Suppose that £DivA = ; and u2(dJ ; µ) ¸ 0 for all µ 2 £CongJ .

Integration dominates non-integration whenever ¸1 = ¸2 = 1, and the expected

gain of integration over non-integration is

E[u(dJ ; µ) j µ 2 £DivJ and u(dJ ; µ) < ¼T2 ] Pr(µ 2 £DivJ and u(dJ ; µ) < ¼T2 )

where

¼T2 =Pr(µ 2 £CongJ )E[u2(dJ ; µ) j µ 2 £CongJ ]

Pr(µ 2 £Cong) + Pr(µ 2 £DivJ and u(dJ ; µ) < ¼T2 )

Proof. Integration induces the ¯rst-best whenever both managers are dedicated

with probability 1. Non-integration results in bargaining ine±ciencies whenever

µ 2 £DivJ and u(dJ ; µ) < ¼T2 . The result of the Proposition follows from the

expression of ¼T2 .

22

Remark 1: If24 u(dJ ; µ) > E[u2(d1(µ); µ) j µ 2 £Cong] for µ 2 £DivJ , then

¼T2 = E[u2(d1(µ); µ) j µ 2 £Cong] and there are no bargaining ine±ciencies in the

non-integrated structure. Hence non-integration and integration have the same

performance whenever ¸1 = ¸2 = 1.

Remark 2: In the setup of Section 3 (and without any further restriction), in-

tegration (weakly) dominates non-integration whenever both managers are known

to be dedicated with probability 1. This is because integration leads to the ¯rst-

best organizational payo® , i.e. E[max(u(dJ ; µ); 0))], whenever ¸1 = ¸2 = 1 (see

Proposition 9 below). However, quantifying the expected gain of non-integration

over integration is a bit cumbersome in this general speci¯cation.

5 Further insights

5.1 When acceptable decisions deliver random performances

Consider a one-organization setup in which the manager has to make a physical

decision d 2 fdA; dJg where dA designates the business as usual decision and dJ is

a decision that requires further justi¯cation. The organizational payo® associated

with decision d when the state of the world is µ is denoted u(d; µ). We ¯rst observe

that:25

Proposition 9 Suppose that u(dA; µ) is independent of µ. The e±cient decision

is made in all states of the world.24

Pr(µ 2 £CongJ )E[u2(dJ ; µ) j µ 2 £Cong

J ]Pr(µ 2 £Cong)

= E[u2(d1(µ); µ) j µ 2 £Cong]

25The result is independent of the probability with which the manager is dedicated. But,

the performance of the organization does depend on this probability since an opportunistic

manager will divert the surplus created by the e±cient decision to his own interests.

23

Proof. Let uA ´ u(dA; µ) for all µ. Clearly, we have that ¼T = uA. Whenever

u(dJ ; µ) > u(dA; µ), decision dJ can be made because it results in a payo® no

smaller than the threshold uA.

However, when u(dA; µ) varies with µ the e±cient decision need not always

been made. For the sake of illustration:

Proposition 10 There are two states µA, µJ which occur with positive probabil-

ity. Decision dA (resp. dJ) is e±cient in state µA (resp. µJ). Assume further that

u(dA; µA) > u(dJ ; µJ). The decision must be ine±cient with positive probability.

Proof. Suppose the decision is always e±cient. Then the threshold is ¼T =

u(dA; µA). But, this does not allow for the e±cient decision whenever the state

is µJ because then the e±cient decision dJ could not be justi¯ed ex post as

¼T > u(dJ ; µJ).

Of course, the randomness of u(dA; µ) does not systematically cause ine±cien-

cies. If in the above two-state example, u(dA; µA) were lower (instead of larger)

than u(dJ ; µJ), then the e±cient decision would always obtain.

More generally, we let £A = fµ j u(dA; µ) ¸ u(dJ ; µ)g, £J = fµ j u(dA; µ) < u(dJ ; µ)g,and uA ´ E(u(dA; µ) j µ 2 £A). It is readily veri¯ed that:

Proposition 11 Assume that u(dJ ; µ) ¸ uA for all µ 2 £J . Then the decision

is always e±cient.

Remark: When u(dJ ; µ) < uA for some µ 2 £J , sometimes the decision must

be dA even for some states µ 2 £J . (Standard ¯xed point arguments guarantee

the existence of an organizational equilibrium under mild continuity assumptions

on u(d; µ) whenever £J is a continuous type space.)

5.2 Varying the set of acceptable decisions

In the above analysis we took the set A of acceptable decisions as given. In this

short subsection, we wish to suggest some comparative statics with respect to the

24

set A. The interpretation of this comparative statics exercise is not at all clear,

as the primary interpretation of A is in terms of familiarity to those in charge of

assessing the manager (hence it can hardly be thought of as a choice variable in the

organizational design). But, in those contexts in which the review board would

be familiar with many possible decisions, it might consider requiring justi¯cations

also for familiar decisions. The following arguments might be of some value for

such cases.

Consider a one-organization setup with a payo® given by u(d; µ) where d

denotes the decision and µ the state. There are two decisions d1 and d2 and

two states µ1, µ2. Each state µi, i = 1; 2 is such that decision di is optimal

relative to d¡i. That is, u(di; µi) ¸ u(d¡i; µi). To ¯x ideas, we assume that

u(d1; µ1) > u(d2; µ2).

We ask ourselves whether the set A should be fd1g or fd2g. As we have seen

above (see Propositions 10-11), if A = fd2g the optimal decision will be made

whereas if A = fd1g sometimes an ine±cient decision will arise.

As a corollary of this observation, we get that if the manager is dedicated

with a su±ciently large probability, letting A = fd2g leads to a more e±cient

working of the organization than A = fd1g. Suppose now that the manager

is opportunistic with a very large probability, say with probability 1. Letting

A = fd2g will result in an organizational payo® of u(d2; µ2) (whenever µ = µ1

the manager will choose d = d1 and he will divert u(d1; µ1) ¡ u(d2; µ2) from the

organization). Letting A = fd1g will result in an organizational payo® of u(d1; µ2)

(that is, decision d1 will always been made, and when µ = µ1 the manager will

divert u(d1; µ1) ¡ u(d1; µ2) > 0 from the organization; the threshold payo® is

¼T = u(d1; µ2)). Since u(d1; µ2) < u(d2; µ2) (because d2 is the e±cient decision

when µ = µ2), we get:

Proposition 12 If the manager is either dedicated with a su±ciently large prob-

ability or opportunistic with a su±ciently large probability, letting A = fd2g leads

25

to a more e±cient working of the organization than A = fd1g.

Remark: For intermediate values of ¸ (the probability that the manager is

dedicated), it may well be that letting A = fd1g dominates A = fd2g.26

Another possible modi¯cation (not considered so far) is to enlarge the set

A of acceptable decisions. But, doing so may be harmful in the presence of

opportunistic managers (because managers are assumed to have entire discretion

over the set of acceptable decisions A). In particular, if the set A is enlarged

to allow for some side-payments, opportunistic managers will systematically take

advantage of this to divert some money from their organization. Even if side-

payments are not allowed for decisions in A, opportunistic managers have no

incentive to choose the best decisions in A. On the contrary, if managers do take

into account that their decisions a®ect the threshold payo® criterion, they have

(weak) incentives to choose the worst possible decisions in A so as to lower the

threshold criterion (and allow them to take advantage of this in good times when

surplus can be generated). Of course, if the manager is known to be dedicated

for sure, giving full discretion to the manager cannot hurt the organization. But,

as soon as the manager is not dedicated for sure, the best set A should not

necessarily consist of all possible decisions.

5.3 On the determination of the threshold performances

The threshold performance in an organization has been assumed so far to be

determined as the expected organizational performance conditional on an ac-

ceptable decision being made in the organization. A rationale for this is that

such a procedure guarantees that the expected performance can be no smaller

than the threshold performance. But, what about other speci¯cations of the

threshold performances? In this subsection we investigate the e®ect of changing26To see this, set ¸ (the probability that the manager is dedicated) su±ciently large, and

adjust Pr(µ = µ2) to be su±ciently small, and u(d2; µ2) to be su±ciently low.

26

the threshold performance in speci¯c environments.

To start with, consider the case of a single organization O; the decision maker

can make one of two possible decisions dA and dJ . Decision dA is the business as

usual decision and delivers a nul performance in all states µ. Decision dJ (which

requires further justi¯cation) delivers a performance µ in state µ. The state µ is

distributed on (µ; µ) with µ < 0 < µ according to a density f(¢) with cumulative

F (¢). The manager is assumed to be dedicated with probability ¸ 2 (0; 1).

Let ¼T be the threshold performance. The manager will choose decision dJ

whenever µ > ¼T , and decision dA otherwise. When the manager is dedicated,

the organizational payo® is 0 (resp. µ) if the decision is dA (resp. dJ). When the

manager is opportunistic, the organizational payo® is min(0; ¼T ) if the decision

is dA, and it is ¼T if the decision is dJ .

The best threshold for the expected organizational performance satis¯es:

Proposition 13 Assume that µ ! 1¡F (µ)f(µ) is non-decreasing on the domain (µ; µ).

The expected organizational payo® is maximized for ¼T = ¼¤ where ¼¤ is the

unique solution to

(1 ¡ ¸)1 ¡ F (¼¤)f(¼¤)

= ¼¤ (2)

Proof. Given the threshold ¼T , the expected organizational payo® is given by

(1 ¡ ¸)[F (¼T )min(0; ¼T ) + (1 ¡ F (¼T )¼T ]

+¸Z µ

¼Tµf(µ)dµ

Easy manipulations show that the best ¼T is positive and satis¯es condition (2).

Remark: When the manager is known to be dedicated, the best threshold

is ¼¤ = 0, which results in the ¯rst-best decision. When the manager is known

to be opportunistic, the best threshold satis¯es 1¡F (¼¤)f(¼¤) = ¼¤, which corresponds

also to the revenue maximizing reserve price in an auction setup in which the

valuation of the seller is 0 and the valuation µ of the buyer(s) is distributed on

27

(µ; µ) according to density f(¢). For 0 < ¸ < 1 the best threshold is in between

these two values.

In the above one-organization setup, the best threshold never lies below the

expected organizational performance obtained when an acceptable decision is

being made (this is 0 in the above example). But, in more complex environments

(in particular those requiring negotiations between organizations), this need not

be the case. For example, in the context of Section 3 if managers are known to

be dedicated with a su±ciently large probability, setting threshold performances

below the expected organizational payo®s obtained when acceptable decisions are

being made would be desirable because it would allow the managers (assumed to

be dedicated) to negotiate more often.

From the perspective on the optimal determination of threshold performances,

the above ¯nding that the lowering the threshold performance may be a good

idea is probably one of the most interesting ones, but our view is that the review

boards need not know the prior distributions of the parameters relevant to the

decision makers. In such a world, it would seem that the reviewing process should

not depend ¯nely of the details of these distributions.

6 Related literature

Grossman and Hart (1986) and Hart and Moore (1988) have developed a very in-

°uential theory of integration based on the hold-up idea (Williamson 1975). Since

di®erent forms of integration result in di®erent hold-up ine±ciencies the theory

provides a rich theory of the ¯rm. In our setup, the decision di in organization

i can possibly be interpreted as an ex ante investment as in the Grossman-Hart-

Moore theory. However, in contrast with the literature on incomplete contracts,

we assume here that the decisions di can be made through negotiations. Our

insights do not rely on the hold-up phenomenon, but rather on the working of

the review processes.

28

Bernheim and Whinston have analyzed the extent to which a contract should

be as speci¯c as possible in a world where veri¯ability constraints impose that

the contract cannot be made contingent on every single observation. They ob-

serve that leaving some discretion (in the form of incomplete contracting) to an

employee may sometimes be good in multi-period setups. In a di®erent vein,

Aghion and Tirole (1997) (see also Holmstrom 1984) analyze the optimal degree

of delegation, and they quantify how much the discretion of the employee should

be restricted as a function of the divergence of interests between the employee

and the employer.

In our setup, enlarging the set A of acceptable decisions can be interpreted as

leaving more discretion to the manager. And the probability 1¡¸M that manager

M is opportunistic is a measure of the divergence of interest between the manager

and his organization. Thus, the insight that when the manager is known to be

dedicated with a su±ciently large probability enlarging the set A may be a good

idea echoes insights in Aghion-Tirole's theory. But, our insights about the cost

and bene¯t of integration (taking as given the sets A of acceptable decisions) do

not have their counterpart in Aghion-Tirole's theory nor in Bernheim-Whinston's

one.27

Rajan and Zingales (2000) have analyzed the cost and bene¯t of integration

based on the idea that a party receiving monetary transfers can use them to im-

prove his bargaining position in future negotiations. This can be interpreted in

the multi-task perspective (because parties can invest in di®erent sorts of activ-

ities) (Holmstrom-Milgrom 1991) or in the in°uence cost perspective (Milgrom-

Roberts 1988). From the bargaining theory viewpoint, the use of monetary trans-

fers for future negotiations translates in imperfect transferability of utilities at the27One possible insight (not much developed above) that may be vaguely related to Bernheim-

Whinston is that if A may possibly consist of several decisions, it need not necessarily be a

good idea to include all of them, as this may a®ect adversely the threshold performance that

de¯nes the criterion of success (see subsection 5.2).

29

bargaining stage, and as Rajan-Zingales shows this imperfection is more or less

severe depending on the distribution of property rights. In Rajan-Zingales' theory

there are bargaining ine±ciencies in the sense that the bargaining outcome does

not necessarily maximize the joint surplus (due to transferability problems).28

But, the channel through which bargaining ine±ciencies occur is very di®erent

from that in the current paper. In particular, in our setup bargaining ine±cien-

cies are due to the fact that review boards are not present at the bargaining stage

whereas in Rajan-Zingales' theory bargaining ine±ciencies arise because of the

inabilities of the parties to commit not to use side-payments to improve their

bargaining position in future negotiations.29

Finally, there is a vast literature that studies how the organizational perfor-

mance is a®ected by the monitoring technology of the decision making (see, for

example, Mookherjee and Png (1992)) and/or by the degree of opportunism of

the management (see, for example, Rajan-Zingales (1998)). The current paper

can be viewed as providing a di®erent and complementary perspective on the

performance of organizations: while the monitoring technology and the degree of

opportunism are set exogenously, we compare the relative performance of orga-

nizations as a function of their degree of integration.

28See also Jehiel (1997) for another source of transferability problems in a public economics

context in which jursidictions bargain over local public goods and takes to attract mobile

citizens.29It should also be mentioned that while bargaining ine±ciencies take the form of a complete

breakdown of the negotiation in our case, they take the form of suboptimal negotiations in

Rajan-Zingales' theory.

30

References

[1] Aghion P. and J. Tirole [1997]: \ Formal and Real Authority in Organiza-

tions," Journal of Political Economy 105, 1-29.

[2] Bernheim D. and M. Whinston [1998]: "Incomplete Contracts and Strategic

Ambiguity," American Economic Review 88, 902-932.

[3] Coase, R.H. [1960]: \The Problem of Social Cost," Journal of Law and

Economics, 3, 1-44.

[4] Grossman S. and O. Hart [1986]: \The Costs and Bene¯ts of Ownership: a

Theory of Vertical and Lateral Integration," Journal of Political Economy,

94, 691-719.

[5] Hart O. and J. Moore [1990]: \Property Rights and the Nature of the Firm,"

Journal of Political Economy, 98, 1119-1158.

[6] HolmstrÄom, B. [1982]: "Managerial Incentive Problems: A Dynamic Per-

spective," in Essays in Economics and Management in honor of Lars

Wahlbeck, Helsinki, Swedish School of Economics.

[7] HolmstrÄom, B. [1984]: "On the Theory of Delegation," in Bayesian Models in

Economic Theory edited by Marcel Boyer and Richard Kihlstrom, Elsevier.

[8] HolmstrÄom, B. and P. Milgrom [1991]: "Multi-task Principal-Agent Ana-

lyzes: Incentive Contracts, Asset Ownership and Job design," Journal of

Law, Economics and Organization, 7, 24-52.

[9] Jehiel, P. [1997]: "Bargaining between benevolent jurisdictions or when del-

egation induces ine±ciencies," Journal of Public Economics.

[10] March J. and H. Simon [1958], Organizations, New York, Free Press.

31

[11] Milgrom P. and J. Roberts [1988], \An Economic Approach to In°uence

Activities in Organizations," American Journal of Sociology, 94, S154-S179.

[12] Mookherjee D. and I. Png [1992], "Monitoring versus Investigation in Law

Enforcement," American Economic Review.

[13] Pigou C. [1950]: \The Economics of Welfare," 4th edition, Macmillan, Lon-

don.

[14] Rajan R. and L. Zingales [1998]: "Power in a Theory of the Firm," Quarterly

Journal of Economics 113, 387-432.

[15] Rajan R. and L. Zingales [2000]: "The Tyranny of the Inequality: An In-

quiry into the Adverse Consequences of Power Struggles," Journal of Public

Economics 76, 521-558.

[16] Simon H. [1955], "A Behavioral Model of Rational Choice," Quarterly Jour-

nal of Economics 69, 99-118.

[17] Williamson, O. [1975]: Markets and Hierarchies, New York, Free Press.

32